45: Using AI for Scalable Prompt Engineering with John Munsell

This podcast episode is sponsored by chiefaiofficer.com, offering training and certification through the International Association of Chief AI Officers. Interested in a new career or leveling up your value in the marketplace? chiefaiofficer.com can help. The following is a conversation with John Munsel, pioneer in scalable prompt engineering. John's the author of the upcoming book, Ingrained: How to Build an AI-First Culture in Your Organization. With a rich background in AI strategy and a passion for helping businesses integrate AI seamlessly, John brings invaluable insights into creating efficient and effective AI systems. In this episode, we'll explore John's innovative AI strategy canvas, discuss the principles of scalable prompt engineering, and uncover actionable strategies for building an AI-first culture within your organization. Welcome to Using AI at Work.

I'm your host, Chris Daigle.

Each week we'll be learning how today's business owners, entrepreneurs, and ambitious professionals are getting more done with smart use of tomorrow's tech. Let's get started.

All right, everybody, welcome to the call. John has been on my radar for, I guess, you know, uh, since we got started. He was very early in when it came to educating, especially on the prompting, uh, paradigm. And he happens to be from South Louisiana, which is where my family's from as well. So I was very interested. Who's this guy from Acadiana that knows all this stuff about AI? And recently, John and I have had the good fortune of getting connected through a mutual friend, which ultimately resulted in a very long conversation prior to us turning the cameras on for this today. But this podcast episode itself. So, John, before we get started, can you kind of just share with everybody what you've been up to? Pre and post-AI.

Yeah, sure. Thanks, Chris. And big fan, by the way. I know you may or may not recall, but we connected— I got it— it was a year and a half ago in our CXO AI Roundtables. And at the time I was like, dog, how did you get that domain name? The best domain name on the planet. So anyway, it's been fun to reconnect with you and love your work, love what you're doing. And, and look forward to being a part of it and also appreciate being part of this. So, yeah, so I mean, 1,000-mile-an-hour tour. I started off in banking and financial services and then the web came out and I thought, oh, this is where the world's going. So I dove in, started a software company in '97. We started building web applications. That was in Tampa. And then in 2003, we moved over here to Lafayette because my wife was born and raised here. I was raised in Baton Rouge, but you know, when you marry a gal from Lafayette, you end up back in Lafayette. So anyway, so we started to grow the company, made the Inc. 5000 3 times, developed into a digital marketing agency. And then, you know, in the COVID zone, we started playing around with AI. And at that time, it was just kind of talking in circles, as I put it. But it started to get interesting. And then when it started to get interesting, I thought, oh, this is the next frontier. So I sold off the agency part of the company and then started the CXO AI Roundtables to try to figure out what problem the market wanted solved, and then how could I build a company around solving that problem. So rather than try to go product-market fit, I went, what does the market want, and I'll try to fit a product into that.

Yeah, yeah.

So, uh, that's what we did. So, um, interestingly enough, I, I have been a fan of, uh, direct mail and direct response marketing for decades. And, uh, one of my— I guess you'd call it mentors— is a guy named Jay Abraham, and another guy, John Benson, who wrote the video sales letter stuff. So John Benson started to get into AI. And so I joined his mastermind group and I'd bought a bunch of his products and I was like, eh, let me just see what he's doing on the AI front. And in it, I met Darby Rollins and Darby had helped John write a bunch of prompts for the product that he was building. And when I saw the style that he was using, the techniques that he was using, that's when it kind of hit me. I was like, oh, this this could go way beyond where he's taking it now because I'm a business guy, right? I work with businesses and I'm thinking like an entrepreneur who's trying to scale up a business. And when I saw that methodology, I thought, if I crank this up a notch and zero in on solving the scalable side of prompting, then I might have something. And as I was going through the roundtables, and talking to people. And, you know, they gradually— they were all getting over the parlor trick phase, is what I call it, you know, where you can ask it a question and ChatGPT responds with something, and they're like, look how much words it produces, right? Um, so, um, that's when I thought, okay, I think that the problem that people are going to inevitably have is everybody— this is so new, everybody's teaching themselves and everybody's got a different methodology. And if, if you're a one-man band, that's no problem. But if you're 100 employees or if you're 1,000 employees and everybody's self-taught and nobody has any discipline, it's a mess, and nobody has any, any idea of what excellence looks like, you have a big problem on your hands. It'll be— it'll feel like efficiency. In the short run, but it'll quickly start to feel like chaos. So that's when I created this process that I call scalable prompt engineering. And the process was cool. It basically, you create containers and variables so that you can swap out these containers and variables, but the prompt itself, the request is separate from those containers and it calls those containers. All right, so that way I can take a a prompt that does operation X, I can swap out a variable and now it does operation Y and I didn't have to write it and experiment and all that other fun stuff. And that means in a company, I could have Susie over here who is writing blog posts and I could have Jane over here who's in HR and they could look at each other's prompt and go, "Oh, you know what? If I swap out this and this, I can use it to write." a job description and other stuff. It goes way beyond that. But anyway, so I started to teach this to people and then I noticed we had a 6-week course and I noticed after 2 weeks that they weren't quite gripping the principles of this. And I said, "Man, I'm kind of a visual learner anyway." And one of the things that I always did in the finance space is I had this ability to take a complicated concept and turn it into a graphic that people could then visually understand. And it was just, yeah, okay, I got to do this again. Well, I don't know if you're familiar with the book Business Model Generation and a thing called the Business Model Canvas.

I wasn't until you introduced me to it.

Okay, well, it's amazing. I saw this guy at an Inc. 5000 conference, I think in 2012. And we blew it up to like 6 feet and we would teach people our business model as we were trying to scale our own business, you know, so they could see all the moving parts, the customer segments, revenue streams, all that fun stuff. And it hit me one night, I was like, man, the business model canvas would be the ideal way to represent what I'm teaching here. And so, it's just, you know, whether it was God speaking or what, it just woke me up in the middle of the night and I I got up and I started to position everything. Yeah, because we had teaching the next day, so I had to teach. So I put it all together, taught everybody that, and the light bulbs just went on for everybody. And it was funny, they were, they were literally saying, oh wow, I get this now. And one of the guys who's probably one of the most brilliant guys I know He's one of the world's foremost experts on TOC, Theory of Constraints, which is a very technical project management process. He looked at it and he's been telling everybody else, he goes, "Once I saw that, everything clicked." And then he was able to take that same canvas and teach his developer how to write prompts our way. And so now they created a help desk essentially inside their software that is a chatbot. And the underlying prompt framework is based on the principles we teach in our training. So it definitely helps. It just gets people to get it and understand it. And at that point, you know, everything clicks. So that's kind of our philosophy, Chris. Our philosophy is if I can get everybody in the organization understanding it from the same level, then adoption goes a whole lot faster and execution is a whole lot more effective and efficient.

You know, I hadn't really looked at it. Sure, the idea of prompt library, right? That's a common term that people are familiar with. And I hadn't really thought about a prompt library outside of efficiency, but I didn't consider the change management necessary if you didn't have that, because it would be, it would be a hot mess with somebody writing. I mean, there'd be no control. So I really like this idea. Um, couple of things. John was our, actually our first guest on the podcast. So I had a chance to listen to how he, how he looks at prompting. And one of the things that I, uh, have taken away from, uh, my interviews with copywriters in particular is my— I'm fascinated because they study language like words and the impact that words have and specific words and all that kind of thing, right, in a sentence. And to have that training go into your prompt building was like a whole new level. It was almost as if they were naturals when it came to at least identifying what the ideal outcome would be from the prompt. And that was a great interview. But secondly, going back to the CXO AI roundtables, first off, I think that's a fantastic idea. If I want business intelligence, let me just host and monetize an environment where everybody that has my answers gets in there and, like, tells me what I need to hear. That's brilliant. But what did you— what were the ahas for you from participating in those conversations?

You know, I think the biggest thing, and it's strange, Chris, because it's now a year and a half later, you read that AI adoption is, you know, at 86% or 90-something-odd percent. Back then it was 0%, and I was like, wow, I can't believe these guys really aren't into AI. And they're basically coming to the roundtable to go, what do we need to know? Yeah. Right, we don't know what this is. What do we need to know? A year and a half later— well, let's just go back a year and a quarter later. I, I was doing a, a live keynote address to a, a convention of— or actually it was a conference for the printing industry in Clearwater. And I'm thinking, okay, it's a year and a quarter into this. And so I asked the audience, how many of you guys are playing around with AI? And I had prepared this talk assuming it would be about 60%. I thought the statistics say 86%. I'm figuring maybe it's 60%. Out of 100 people in the room-ish, you know how many raised their hands?

Less than 10.

4. I— yeah, great guess. I would never, never have guessed that. You have 4 raise their hands, and I'm like, oh my God, I have just prepared the wrong talk for this audience, you know. But fortunately, I was, I was teaching them about the canvas, and, and it went over hopefully fairly well because I got, I got some, uh, business out of it. But it, it was stunning to me that, that, um, late in the game people were that far behind. And then here we are still even farther along and I can ask that same question. Like, we had a webinar— now this was for Southeast Asia about a month ago— and we had, I don't know, 30 people on the call. And once again, I was asking, how many of you guys are using AI? And you just have a couple of people sheepishly raise their hands. And I'm like, wow, why aren't y'all in this so much? So anyway, that was— that's, that's the, the fascinating thing to me. Is that I don't know whether it's just the small businesses, the medium businesses, or the what, but the degree to which they're not fully immersed is stunning to me.

You know, I had this conversation with somebody yesterday. I was recording a podcast and the same statistic came up, the Microsoft study that showed like a number that I don't believe of adoption. And I think what it might be, it might be They're assuming that if, okay, you're a Microsoft client and you've bought Copilot, you must be adopting. But we know that people aren't trained on it. They're not using it. Um, so a couple of things. One, I'm encouraged by that because when we launched our business, we launched our businesses about the same time, you and I, I thought we were late a year and a half ago. I thought we were late and, you know, starting in 2023, I was like, Okay, it hasn't happened yet, and I get it. People are still kind of getting their feet under them. But boy, we're coming up on the end of the year, 2024. They're going to— it's going to be everywhere. And it hasn't happened. Right. So we're still early. And I totally agree with you that if I were to go and present to an audience of business professionals, say, who's using it? It's, it's going to be limited to maybe some ChatGPT, but but not sophisticated thought behind, because it's not just the tool, it's you thinking about how to use the tool, how to integrate the tool, or how to become, as Ethan Mollick called it, that cyborg when he talks about that Boston Consulting Group study, like the seamless use of it, like you and I, right? It's seamless. We don't even— there's no business consideration that we make that isn't informed by, oh, can, can AI do it? Can AI help me with it?

Right.

So, okay, well, I'm glad to hear that because that means that we're going to be even more ready when it happens. Beautiful. So I would like to get into, once you had this aha about what was possible with the prompting, and you said the way he was doing it, were you talking about the way Darby was doing it or the way John was doing it?

Darby, therefore John. Yeah. So the, the, because Darby essentially originated John's first series of prompts. And what I thought was so really intriguing that again, I think Darby—

And guys, we're talking about Darby Rollins from GenAI Academy.

GenAI University. GenAI University. Correct. Yep. Yeah. And he's got a— Market summit next week that I'm going to be a speaker on. But yeah, it's like he was doing it to solve a problem for John. And what I saw from that was this, if you ratchet it up a notch, can solve a lot more problems than just direct response copywriting. And that's where I started to take it from a business perspective. How do I, you know, first of all, how do I teach my employees how to use this at at scale, how do they repeat stuff, how do I make sure everybody's doing it the same way? 'Cause it's like you said a second ago, everybody's talking about, oh, you gotta have a prompt library. But if you don't understand how Bill created his prompt and it's just a paragraph after a paragraph, you're just like, I don't wanna do this. Let me just create my own prompts in my side of the library. So now it's not scalable, now it's not really shareable because nobody's doing it the same way. But when I saw Darby's stuff, I was like, okay, if I do this and I kind of marry it up with some of the principles that Claude, Anthropic teaches in their framing, so where they use delimiters, but a lot of people are talking about delimiters. The problem that I see with the way they're using delimiters is they're, I would call noisy. Things to be clean and simple. And a delimiter simply just designates what's inside of what we would call a container. You know, it says, here's, here's delimiter that says it begins, and here's the delimiter that says it ends. And Claude, they tell you to use the equivalent of XML, you know, where you're— or HTML, where you're opening tag and you're closing tag. You don't really need to be that complicated. It can be simpler than that. But as long as it knows what the beginning of the container is and the end of the container, you're in good shape. So when I was looking at John's, John's never really had an ending of the container. It had a beginning. And I was like, hmm, that's interesting. But he only had essentially one container and a bunch of variables. And I thought, okay, if I had multiple containers and multiple variables, then my prompt is just simply the request and I tell it what to do with all these other containers. So instead of me having this giant paragraph that tells the AI all sorts of stuff and I end up with what we call prompt conflict, where you have a sentence over here that conflicts with a sentence over here because you've made this giant story. Yeah, yeah. Then you got problems. But when you bracket it up using these delimiters and containers and variables, now all of a sudden you eliminate prompt conflict and you have a very efficient method of telling the AI what to do and how to execute. And, and somebody can come behind you and see the structure and they know exactly what's going on in the prompt and they know exactly where things are potentially in conflict and how to fix it. And that's where it becomes scalable, right? That's why we call it scalable prompt engineering, because if everybody in the organization knows it from the same methodology and has the same a frame of reference, which is what the AI Strategy Canvas gives them, then the use becomes a lot more efficient, effective, proficient, and prolific. There's a bunch of words for you. Yeah.

I love it. And it makes perfect sense. It's how do you train somebody how to use that? Use the scale? Like, yeah, let me ask so that I can better understand it. Where are they typically stored in a client environment, like in a Notion database or something?

We use Notion. Yeah, we teach them how to use Notion. There are other tools that can store databases or store prompts. The key is to have a database that's going to give you fields that you can populate. Um, like, I would want to know what stage of the game the prompt is. Is it complete? Is it shareable? Is it under construction? Are we testing it, you know, I want to know that. I want to know, is the prompt a prompt or is the prompt a container or is the prompt a set of variables? I want to know, is the— is in the database itself, like we have 8 follow-up prompts inside of one prompt, you know, the position for that. So in other words, if I have a prompt that's going to generate what we call a Newsjack post, that initial prompt might have 2 containers and 8 variables in it, but then there will be follow-up prompts, which would be what we would call a prompt chain, right? So we're gonna do this request and that gives the AI some information, and we're gonna ask it to do this, and then we're gonna ask it to do this, and we're gonna ask it to do this. And that is all, you know, if somebody says, oh, I need to go do a Newsjack post, they just pop that up, and then they see the exact sequence they need to run and they can paste it in there. Now, those prompts also can become the custom instructions inside of a custom GPT or inside of a project. And so now instead of you having to cut and paste and cut and paste, you've got a project that knows what it's supposed to do and it has all that embedded in it, right? Smart. Yeah. So it becomes more simplistic. So as AI progresses, you know, somebody asked me the other day, they're like, hey, you know, as AI progresses, they're not going to need prompts. I'm like, I, you, you had it all wrong. I mean, like, just because you hire a smart employee doesn't mean you have to— you can't onboard the employee. Yeah, you need to teach the employee how to do his or her job, right? And so we still need a method to teach AI how to do its job and to do it effectively. And there are, there are lots of things that come up every single day that you that aren't repetitive. But if you think AI first and you know how to structure a prompt the right way, then you might be able to go to your library and pull bits in, but do other stuff and get you there faster. So that's, interestingly enough, that's what the book that I'm writing is called. It's called "Ingrain: How to Build an AI-First Culture in Your Organization from Strategy All the Way Through Execution." If you're enjoying this episode and want to learn more about how to start using AI at work, We've made it easy for you.

For just $1, you can have full access to the Chief AI Officer community, which will give you additional training, custom software, daily training calls on AI tools, using AI automations, getting more from your ChatGPT sessions, and the business of being an AI consultant. Simply go to chiefaiofficer.com/insiders to accelerate your AI journey now.

Back to the episode. Hey, so John, before we, before we go past that, when are you, when are you planning on releasing the book?

Oh, well, as soon as you get finished telling me it's worth it.

So everybody listening, we'll be making an announcement for sure to support John's book launch, but this is going to be the type of book that isn't just a fluff theory kind of thing. This is going to be very actionable stuff. So I would encourage you to Follow the messaging and when it comes out, grab a copy.

So, anyway, I didn't mean to interrupt you. I'm hoping, Chris, that it'll be out by the end of November. My goal was to have it out at the end of October and I failed miserably. But it walks people through the strategy canvas because that was kind of your original question is how do you get people thinking this way? The strategy canvas is the thing. There are 9 blocks in the strategy canvas and it makes people think through methodically what do I do I need to give the AI in order for it to give me the most effective output from the first shot? And then what do I need to do to think through AI strategy as a business? And you go through those same 9 blocks to help you think through, okay, what kind of initiatives can we have here? And then, and in the book I teach this, you look at the canvas, you come up with some ideas for some initiatives for your overall strategy, But when you're defining strategy, you have to think of the 3 strategy types, 3 AI strategy types, right? So, one would be an innovation strategy, one would be a customer engagement strategy, and one would be an efficiency strategy. So, you use those 3 strategies to help you think through, okay, what do we want this particular project to be? Do we want it to touch customer engagement or do we want it to be all about efficiency? You define the engagement or define the initiative, and then you go back to the canvas because now you have potentially people from other areas of your company that need to be involved in that initiative. So, it might be that you need to get legal involved, you might need to get IT involved, you might need to get HR involved. And now, they look at the canvas again from a different perspective. But this time, they take the initiative and they deconstruct it is what we call, and I outline this in the book. We deconstruct it. Into those 3 strategy types. So, it comes at you as maybe a customer engagement initiative, but now when you look at it and say the customer engagement initiative was a chatbot for the website to handle technical support, whatever. Now you look at that same initiative and you deconstruct it into those 3 strategy types because that initiative might have some innovative potential. That initiative might, of course, have some customer engagement potential, but it also will have some efficiency potential. And so now you're separating it out and saying, okay, what do we need to be doing to look at this to be more efficient? And then what do we need to do to make it more innovative, something we haven't thought of before? So now it becomes this big brainstorming tool to really flush out that project. Okay? Now, once you've got it all flushed out and every department now knows what they need to provide, because the canvas will tell them, oh, you know, who are we talking to? Who's going to be the beneficiary of this? That's the target audience, right? What other context do we need to provide? What are the resources that we need to provide? What's the style, the brand voice? What do they need to know about the company and our products and services? And then are there rules that we need to try to maintain? Are there regulations that oversee this? And then you finally get into the request. All right. So now you take that same canvas and the people that are in charge of using this tool or programming it on the backend, they're gonna use that same canvas now to do the prompt engineering. And that canvas teaches you what those buckets should be or those containers that I was telling you about, how to create an efficient prompt structure. So now you've got potentially somebody who's building the automation who needs to know how to create the custom instructions, But then you also have somebody on the front end who's going to be using it who doesn't know how to do all that other fun stuff, but they need to at least mess around with Claude or ChatGPT or Perplexity to get something going, right? It teaches them all that. So what the benefit is, is that everybody who touches it knows what the canvas is and knows how to think in terms of a more holistic approach to AI. And they know what all the ingredients are. You know, you'd have to see the canvas to fully understand the, uh, the whole tool. But do you have— that's how it works.

Do you have a visual that we could include in the show notes?

Yeah, yeah. I don't know whether you want me to share the screen, but I can pop it up on here if this is also video. But yeah, okay. So this is the canvas itself. Now what we have here, let's see if I go over here, you see they're all blank. This, this is just kind of our primers to teach people a little bit about what's—

this is all covered in the book.

Yeah, this is all covered in the book. Right. So we, we start now. And again, this is, this is— people always ask, how come it's going right to left instead of left to right? How come it's going counterclockwise instead of clockwise? I explain all that in the book. There's a, there's a very specific structure of this. If you're thinking in terms of accounting, you have debits and credits, right? Debits on the left, credits on the right. What happens when you increase a credit?

Debit is—

you're increasing income or an asset, right? What about a debit? Debit's going to be a liability or an expense. So it's the same thing. We got expenses over here and we got assets and, and revenue over here. All right. At the bottom, if you were looking at the business model canvas, this would be income and this would be expenses. All right. So the request is really the final step. And that's the problem. Most people just dive into the request and they're going back and forth and back and forth and back and forth to try to ultimately figure out what am I missing? Why isn't it working? We start with the target audience. Who are we creating value for? Who are we talking to? Who's the beneficiary of this AI? Project or prompt. Then we want to know what's it need to know about our company? Does it need to know a little bit? Does it just need to know the name of the company or does it need to know our history, our values, you know, on and on and on? Same with the products and services. Are we talking about a specific product? Are we just giving it some global information about all of them? What kind of detail do we need to know about the value we deliver to the customers and the transformational benefits we provide?

Each of these are a container?

Yeah, they could be, or they could be like in our case, if we're building a GPT, this would be several documents. This would be a solid document. This could be several documents, a persona for the 3 personas that you're targeting. It just depends on what you're trying to execute. In a prompt, it could just be a stack, a container or something, right? Okay. The context is what else? Do you need to give it? Because all these are essentially context, but this is more unique context. Like if I'm going to write a blog post, the context might be current events, might be my thoughts on the subject. Like when I'm writing the book, I have all of this stuff in big documents inside the GPT, and I'm constantly giving it my thoughts to guide it where I want it to go with the book. The role is what's the role the AI is actually playing? What are you asking it to play? And then there's the style and the brand voice. We use a calibration system rather than a descriptive system because I want to know, you know, if I tell it to use a little bit of humor, it's how does it know what a little bit of humor is? So we literally gauge it from 0 to 10 so that it's a little bit better. Now, the problem is these two will get into conflict with one another. And unless you've experienced it, you don't know that actually happens. But depending upon how you name your role, it's going to assume a style and a brand voice. And that's why the calibration method is more effective, because I can calibrate up against whatever the role is, or I can just get rid of the role entirely. Typically, you only want to use a role for a certain exchange with the AI, and then you want to drop the role. Okay. Because you don't need it the whole time. Yeah. And then what other resources are we going to give it? Are we going to tell it to go scan a URL? Are we going to upload some documents? Are we going to give it a database, etc.? And then on the rules, depending on whether you're talking about strategy, the rules could be governmental oversight, could be compliance rules. If you're down there at the prompt execution level, the rules might be, I don't want you saying, I hope this email finds you well. I don't want you saying, skyrocket, that kind of stuff.

Game changer. Yeah.

Game changer. Right. Right. Journey. Your AI journey. If I hear that again, I'm going to shoot myself. And then finally, the request, and that's where you're telling it what to do. But in your request, you frequently have to tell it to reference these other blocks that you have given it, right? Or you tell it in the knowledge base that you got this document, right? So your request is that, and your request also could tell it, I need you to go step by step. I need you to review stuff, right? So you can do a lot of fancy stuff. With the request. So that's what the canvas does. And again, you use it in 3 different contexts. One, to define strategy and have your team think about everything that you do as an operation. And now what, what are the parts that are going to be called upon to develop this strategy? And then you would use it also in terms of the initiative that you define. And what are the moving parts of the initiative? Who do we need to get involved in it? What are the departments need to be present as we build this thing out? And then finally, you know, for the prompt execution, you use the same framework to develop the prompt.

So this is interesting where most people, yeah, most people would start with, like you said, the request. And as a result, there's going to be a lot of iteration that has to take place before they start getting into ideal output. With this, you do the clarity up front so that your output from the initial request may be already your ideal output.

Yeah, exactly. Exactly. It takes maybe a little bit more time up front.

Right.

But, but what happens is you nail it the first time. Yeah, exactly. And everybody else knows to do this. And so that helps you structure it correctly. Like, for instance, we taught 10 LSU professors last month.. And one guy was doing grant research, right? So he applies for these grants. He was applying for a multi-million dollar grant and he had been using AI. He went through LSU's AI prompt engineering course, so he had a little familiarity with it. And he said that after he did it our way, he got what he wanted instantly, like the first try. And so instead of him going back and forth and back and forth for hours, he's like, dang, this nailed it. Yeah, not, not only that, but he got a compliment back, uh, when he submitted his grant proposal. He was submitting it with 9 other universities, okay? And, and these were bigger names than LSU, okay? So he submitted his piece, and one of the professors at essentially a competing university, shot him an email back and said how articulate he was, how the, the story was amazing, how he made his case. And he said she— or that this gal said this was the best grant application she'd ever seen. And so he had to, you know, call me up and tell me. He goes, I got to share this with you. And he did that in one of our office hours where he was like reading her email back. Now this guy's primary language is not English. So he's got a very heavy Spanish accent. Yeah. And so for her to say this was really articulate was really saying something, but it really helped him. And he built an entire GPT to help him write grant proposals now. Yeah, he's nailing it with this. So yeah, that's, that's, that's it.

You say that it takes a little more time up front. What is that? I mean, we don't have to necessarily talk about outliers, but the typical opportunity for a company, a client company of yours, or what is— how much investment in time and resource is required to get that first version of the scalable prompt?

Well, I like to look at it this way. The course used to be a 6-week course. A lot of people can get through it in 2 or 3 weeks. Once you get used to doing this, it's second nature and your prompts just start that way. Nice. Right? But in terms of building a GPT or something like this, I, man, it just varies depending upon how complex your GPT. Like we're working with a client in New Orleans right now and she's in the construction business and she's building a GPT to help her do estimates. And we were meeting with her literally this morning going through her GPT and the request she did for it to go and dissect all these bid requests that come in and stuff. So she's been working on it for a couple of weeks because that's a really complicated thing. And she's also trying to ask ChatGPT to do stuff that it can't do. I was like, look, what you really need is several GPTs if you want to do it the cheap way. And then you're going to have to learn how to call one inside of the other. On the other end, if you want to do it a little bit more expensive, then we need to use a tool like Make or Respell or Cassidy, and we'll create an AI agent that goes out and does a lot of this stuff. She was thinking that she could get it to go out and scour the web for potential opportunities that fit their mold, and then it would also determine whether or not this was near a waterfront and would have to conform to certain maritime rules and all that. I was like, dude, you can't get it to do all that in one prompt. Yeah. But, you know, hey, at least she was thinking that way. But she had done a really good job of working through the canvas and knowing what she would use this for to create a big document that would sit as a resource. You know, she created a big document about her company because, you know, when you're doing those kinds of things, you need to know your tax IDs, your permit numbers, blah, blah, blah, blah, blah. So she was creating all those so that it wouldn't have to constantly ask her for that information. That's just one example.

So, okay, a company, they're listening to this and they're like, let's do it this way. Who's the, the typical person that owns this in a company? So, um, you talk about AI in general, or more like, more like specifically the scale? Like, who's the person who's going to say, okay, I understand this canvas, and now I'm gonna identify the areas in the business where we're going to start building these scalable prompts to support that activity in the business?

Well, you know, in, uh, in most organizations, they're going to have to probably assign or hire a chief AI officer, or they're going to have to create some sort of an AI council and, and appoint somebody over that. But there, there needs to be somebody that ensures some sort of conformity. It's like half of me wants to say it would sit with IT, but the other half of me thinks it probably shouldn't because I think IT would have a tendency to put it in too many constraints and marketing would take too few constraints. And that's why I think having a chief AI officer makes a whole lot more sense where that person can be a lot more broad and yet still bring some regulation to it and some authority to it, right?

So, are there any industries or types of companies that tend to be, like, I guess, get the most benefit out of the scalable as compared to the dynamic, hey, what do I need? Oh, let me do this.

Yeah, so honestly, everybody can benefit from the scalable side of it because once you get the handle of this, the way you write prompts is faster and more efficient. Okay, so like I've taken prompts that were, I don't know, 800, 900 characters and I reduced them to 50. And I've done the same thing with words. I've seen really massive prompts.

Because you're able to use all the variables and the containers and that sort of thing?

Yeah. Exactly. Instead of you writing it out with a giant paragraph, you literally just do really small words and it's way more abbreviated. Yeah. And so it's faster. And, you know, you think about it like I took, I think one prompt was 250 words that I saw. It was a lot. And there's a lot of room for conflict, a lot of room for the AI to mess up. And I reduced it to about 25 words and I got better results, far more effective, far more efficient, no conflict on the prompt side of it. And then once you do that, like I said, you start to realize, okay, I'm going to ask AI to do something. Oh, wait a minute, I don't need to go, I want you to do blah, blah, blah, blah. You just say search blank. You know, like for instance, I don't have like, oh, okay, I'll give you an example. If I want to do a competitive analysis, it's a multi-step process, right? There's a whole lot of stuff I want to get out of it. I've seen prompts that, you know, go into this big laborious thing. Oh, I'm going to do all this stuff. And 90% of it AI can't do, but it sure sounds good if you're the YouTuber, you know, but AI can't do any of that stuff. On the other end, if you're, if you want to just say, look, how do we compare to this competitor? Then you would put a block that says URLs and you would have like maybe the 5 URLs in the competitor's site, but your, your custom GPT would already know everything about your products, your services, your value proposition, et cetera. And you just say, analyze these URLs. And compare it against our product. But the prompt itself, you know, I'm talking in paragraphs, but the prompt itself would just be URLs, paste, paste, paste, and it would say scan URLs and contrast with our products and services. Tell me where we have a competitive advantage and tell me where we're coming up short. And that's it.

Those custom instructions behind the scenes would have all of the context.

Instead of saying, oh, compare me to, I don't know, man, compare our course to Stanford University's AI for Business Leaders course. And it's not going to know. It'll make stuff up. But if you gave it URLs, told it to analyze the URLs and do all that stuff, now you're keeping it from hallucinating.

Right. So let me ask, are, are these typically, you've mentioned, uh, custom GPTs a few times. Is that typically the best place to, to build out the canvas and then load that into a custom GPT for that activity?

Um, well, you're gonna do that, um, in, in any custom GPT or Claude project, it needs a knowledge base. Yeah. I use the canvas to help me think through how to compose the documents in the knowledge base. First of all, what documents need to be in there? So I'm going to go through the canvas to say, do I need to address the target audience? Does that need to be part of my prompt or should it be more voluminous and be a core document inside the GPT? Do I need to tell it what my style brand voice is with 8 variables, or do I need to just use a single variable? Do I need really deep persona documents, or do I just need to say our target persona is CEOs of startup companies, and here are their pains, frustrations, and here are their goals and preferred solutions, right? So it just depends on what you're trying to execute. But the point of the canvas is to make you think through it all and say what needs to be part of the prompt and what needs to be a core document if I'm going to build the GPT. That's really the key.

Well, I'm very excited. Yeah, I'm very excited by getting my hands on the book. But in the meantime, I'm going to start winging it just based on what I've learned from this.

Yeah, I mean, that's the beauty of the canvas. Yeah, the beauty of the canvas is if, and on our website, I believe if you go to bizuka.com/ai-strategy-canvas, you can see all about it. You can request a copy of it and download it, but there's also a little 18-minute video that was me talking on stage to, ironically enough, that printing deal where I explained the canvas. A little bit. So at least you get an idea, you know, more about what is involved with the canvas. But in the training, we teach you how to use this stuff to create these really big, rich documents that we're going to be using AI to solve for certain variables that we want to, uh, learn about. Like, if I want to learn about my, um, persona, I'm not going to just tell ChatGPT, hey, create a persona on you know, a CEO of a startup company, I'm going to give it a whole lot more stuff that I want it to solve for, you know. What does it— so we teach you how to do all that.

What does the training look like? You said it, it's a 6-week program, but people can do it faster. Is it cohort-based? Is it— how does somebody get trained up on this?

Yeah, great question. So it, it used to be cohort-based. Now it's kind of open enrollment. Um, so it's a, uh, it's all video-based training, and we have this thing we call the AI Skills Builder, which what we're trying to do is make sure that a company— okay, the Skills Builder, think about it this way. The Skills Builder is a suite or a series of trainings for a company, and it's skills track based. So I can take 20 employees and then I could look at the skills bidder and say, okay, I need 3 people to go through the marketing skills track. I need 2 people to go through the HR skills track. I need 1 person to go through legal. I need 2 people to go through customer service. So there are 10 skills tracks in it. But right now we have open enrollment for AI for business leaders, which is a skills track. We're getting ready to have open enrollment for the AI for educators where we're teaching university professors and high school professors how to use AI at scale for their own, uh, simplicity, to simplify being a teacher. Yeah, right. But so, so that's how it works. So, but they're, they're video-based, but every Friday— and we'll probably open it up to Tuesdays and Fridays, um, when we do the, uh, the AI for Educators coming up here in, in 2 weeks— but we have these things called office hours. So you're, you're, first of all, you're, when you sign up for the course, you're in an online community. So you can ask questions in the community and we'll answer the questions. The course will be dripped to you every 3 days. Some new stuff will be opened up because we don't want you binge watching like it was, you know, the latest episode of The Simpsons or something. But, um, so we don't want you to do that. But, um, when we have the office hours, you should have learned, you know, certain aspects of it and come to the office hours with questions. The goal of each skills track is at the end you have a capstone project that you have to present to us. And the capstone project is you building a custom GPT or a Claude project that solves a specific business problem for you. So like I said, we had that one gal who's building an estimator GPT and she has to present next week. We have another guy who He's already presented to me one where he is, he probably files for, I don't know, 20, 30 patents a year. And he's using his to do patent research so that he can save on legal fees. Nice. So he's built this amazing tool to, he's in the oil and gas sector. So he analyzes all these oil and gas patents to figure stuff out. He's a genius in his own right. And then we had, you know, teachers building GPTs just to help students get through the, the course syllabus. Yeah. When I was in school, a syllabus was 2 pages long. Now they're 60, 70 pages long. I'm glad I'm not in school. Yeah. I'd need a GPT to do that too.

Yeah. That's very cool. So for the ones that are going to be open enrollment, those are now are coming online regularly. Are there— but what if somebody wanted to do the track on one that wasn't open enrollment, like the marketing or like the customer service or something like that?

It will probably be open enrollment in the next month or two. Okay. We may do a live cohort. We were just debating. It depends on what the demand is. Like, for instance, legal. We may do a live cohort on legal, and I might bring in legal experts on that as well. Yeah, but we haven't quite decided yet. But we have, like I said, there are 10 skills tracks and pretty much all of them are ready for us to do open enrollment with the exception of legal and HR, maybe operations. The only thing we're trying to do is make sure that we have the right practical applications modules. So there, there are 3 foundational modules that everybody goes through. And that's safety, security, and ethics, the AI strategy canvas, and scalable prompt engineering. And then they go into their skills tracks, which is about practical applications for their job, right? So, practical applications around marketing or around sales or around HR, right? And there are 4 practical applications things, and each one of those modules is multiple lessons. And essentially, we're teaching you how to go through some of these blocks in the canvas and expand them to really rich documentation that you're going to be using. And the final module is what we call AI Automations 1, and that's where you're building the GPT. So all of the practical applications modules are giving you all the information that you need to now deliver this GPT so that it's very powerful and very robust. That's the way these things go.

I love it. I can't wait to get my hands on it, man. This is awesome. Well, John, thank you so much for taking time out of your schedule to chat with me about all kinds of stuff prompting stuff. Where's the, where's the best place if, if I like to follow people on social, particularly people like you, to see what they're thinking about when it comes to AI? Are you doing that on any of the platforms?

Yeah, no, I'm, I'm on LinkedIn, so I believe I'm just JW Munsel. @jwmunzel, whatever it is with LinkedIn. You know, they got the long URL. I'm on Facebook occasionally. I don't really use it for business. I just am there. Yeah, maybe. I don't know. I'm just— I don't know. It's becoming too noisy for me. Yeah. So you're like me.

All you talk about is AI anyway.

Yeah, right, right.

You're not interested in the stuff on Facebook, but LinkedIn seems to do a good job of feeding me.

Yeah, exactly. I'm not really interested in what my buddy had for a bottle of wine for dinner the other night. But, uh, but yeah, and LinkedIn is really where I pretty much hang out. Bazooka has a LinkedIn, um, profile as well. Um, probably the best way, I think, would probably just go to bazooka.com. We have a blog there. We publish a lot of content there. I'm on a lot of podcasts like yours. And if you— here's another thing I'll do for you, for your handy listeners, Chris. Nice. And this will put, you know, the, I guess, the fuel underneath my seat. If they go to bazooka.com/ingrain, I-N-G-R-A-I-N, and they tell me that they heard about this through Chris Dagle's podcast, um, Using AI at Work, then I will send them a copy of the book for free.

Heck yeah.

Okay, I may charge shipping, sure, uh, but I, I'll probably have about 100-some-odd copies that I'm gonna give away for free. But I'd love to do that for your listeners because I, I'm a fan. I love what you're doing, love who you're reaching, and love to help. Awesome.

Well, I'll make sure that all of this is in the show notes., and I'm looking forward to, reading that manuscript as well. John, again, thank you so much. And everybody, we'll see you on the next episode.

Thanks for tuning in to Using AI at Work. Don't forget to subscribe for more conversations about how to use AI at work. And a special thank you to our sponsor, Chief AI Officer, for empowering businesses with AI education and training. Visit their website for a free AI readiness assessment and AI strategy guide to help you get started using AI at work. That's www.chiefaiofficer.com. Special thanks to our producer Evan Desaulniers for making this episode possible. Follow us on Twitter at the handle @UsingAIatWork and visit www.usingaiatwork.com for free resources to help you harness AI in your role.